An intereswting co-processor chip

Are you familiar with the phrase, "You had me at..."? Well, I lost all interest at the point where it stated "uses both I2C and SPI interfacing". Even if it had used USB2, exactly the same thing would have applied.

I'm not quite sure what the point of such a co-processor would be when cursed by such glacially slow interface options. Others have suggested possibly beneficial uses with ancient 8 bit systems which were probably the original target market for this chip. A RasPi is most definitely not a suitable target system for this chip.

It *also* depends on the quality of the capacitor, or even just the electrolyte used. There are well-known cases from the last few years of electronic equipment that does not last as long as it should have because somebody has tried to save money.

Doing a quick google for

samsung capacitor

and

dell capacitor

illustrates this quite nicely.

That is true, but that was in 2004. Now another generation of equipment already has failing capacitors that is not due to that problem but only to the higher loads on capacitors in modern equipment (combined with using capacitors with insufficient rating of course, but that is not comparable to the 2004 industrial espionage incident)

What that incident showed very clearly was that all the people who "only bought class-A brand name equipment because they user higher quality components than the others" had been wrong all the time. The failure rate because of those inferior capacitors was the same if not higher in brand-name equipment. Apparently they shop for the cheapest source like all the others, and don't do any additional quality checks.

If you guys are interested in a fast coprocessor chip, you might want to consider the GA144 from GreenArrays. It has 144 small CPU cores each running at up to 700 MIPS. Floating point would need to be done in software emulation, but with that many MIPS it would not be a major limitation. I'm not sure if there is a floating point library available or not.

The chip is $15 or so and uses very little power when not used. Being asynchronous the processor power scales with usage to a max of around 1 Watt with all processors cranking at 100% (which pretty much never happens in any real app).

This might make an interesting coprocessor for either the Arduino or the rPi.

In the case of Samsung, there are reports of problems well past 2004 (like 2008, at least).

What those incidents show is that there are clearly differences between capacitors.

Having been an electronics designer, I'd say that that is a bit simplistic.

ANY elecrolytic is a suspect component and were pretty much banned from all mil spec equipment since the year dot. OK power supplies needed em but the smaller values were (very expensive) tantalums.

You cant justify that sort of expenditure on a consumer item. The use of switched mode power supplies and then aluminium electrolytics exposed a pretty bad feature of them. A high ESR and high speed ripple made them get hot, and lose the electrolyte, raising the ESR still higher..

It wasn't especially a 'bad supplier' or a 'bad batch' - it was the whole lot of them and it just so happened that some identifiable ones were the ones that just were not quite as good as the others.

Constructing for lower ESR has pushed the problem away a little, but its still there as the weakest link...

No! You don't say!

The 2004 incident was traced back to a manufacturer that had done industrial espionage to get the formula for some chemical used in the production of electrolytics, and did not notice they were receiving incorrect data.

They offered very attractive prices, and everyone bought their products (including the top brands). Then the premature failures were noticed.

When the top brands really had bought the top quality components, they would not have fallen into this trap.

Strictly speaking your "no bad supplier" theory is not correct. Yes, electrolytic caps are a weak link in the reliability of most electronic gear, but they are not so bad as to impact the reliability of most consumer gear that has a limited life due to obsolescence anyway. I remember back in the 2002 time frame (if I remember correctly) PC motherboards were failing left and right. The problem was eventually traced to a supplier of the materials for the electrolytic caps who was taking shortcuts on the quality. These devices would fail in the 1 to 2 year time frame rather than usual 5 year plus time frame.

But you are right that electrolytics can create reliability problems in electronics. I remember talking to some who repaired old tube TVs and radios. They said that if it wasn't a tube, the next thing they would look at were the electrolytic caps. So the problem has been with us for a long time.

Back in 1995 I worked for a company which repaired stage lighting and the associate control equipment.

The workshop mantra was "It's the caps, and if it's not the caps it's the component with the most legs"

Which was true more often than not.

-Paul

In my case with high power audio

Its the power transistor, if not the power transistors the ones that feed them'.

So true was this, that a girl of zero technical knowledge, but skilled with a soldering iron, was employed full time to change those out and put all the removed transistors in a pile, and pass unit back to test where someone with a slight clue would pass 97% of them as fit for return.

Once every few days, the pile of removed transistors were tested, and the obvious blown ones thrown in the bin, and the remainder went back to be recycled into more repairs.

The moral being every design has a weakest point, but it ain't always the same point.

return and fix the unit was a 1p polystyrene capacitor, which had melted during flow soldering and shorted subsequently.

Job lot picked up cheap...

Apart from ageing valve radios (like 40 years plus) , until the SMPSU thing, I had never had any issues with electrolytics at all.

It is worth considering that while the co-processor is doing its thing, even slowly, the main system is free to get on with something else.

this of course requires careful program design if it is to be of any use.

How about adding another rPi as a coprocessor to the first? Then you can speed up *all* the work, not just a bit of floating point fluff. BTW, in the case of the rPi, the coprocessor will run much slower giving the rPi a *lot* of time to get on with other stuff. lol

So, this thread is all about adding FloatingPoint hardware to rPi? Like adding a jet-engine to your classical wheel-barrow? Like the boy's in the 50's-60's souped-up their cars: because the GM engineers didn't know how to do it properly?